{
  "batch": "read1-034",
  "plugins": [
    {
      "pr": 0,
      "name": "done-bell",
      "does": "Notifies when Hermes finishes a turn or awaits approval, with a Desktop status chip, settings pane and CLI bell.",
      "useful": 4,
      "reliable": 3,
      "polish": 3,
      "creative": 2,
      "beyond": 4,
      "total": 20,
      "best": "Small, credential-free lifecycle integration includes mute controls and local tests.",
      "worst": "Setup uses an author-specific Windows path and manual Desktop copying; tests do not exercise real OS notifications.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 0,
      "name": "homeassistant",
      "does": "Connects Hermes to Home Assistant through four device tools, filtered WebSocket events and notification delivery.",
      "useful": 4,
      "reliable": 5,
      "polish": 4,
      "creative": 3,
      "beyond": 5,
      "total": 26,
      "best": "A genuine gateway platform with scoped secrets, bounded requests, reconnection, entity restrictions and extensive policy tests.",
      "worst": "Requires an existing Home Assistant installation; device permissions remain broad unless the operator configures entity restrictions.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 149159,
      "name": "understudy",
      "does": "Replays decisions from local Hermes history on candidate models and reports judged takeover rates, misses and projected costs.",
      "useful": 5,
      "reliable": 4,
      "polish": 5,
      "creative": 4,
      "beyond": 5,
      "total": 28,
      "best": "Personalized model evaluation combines read-only history, blind comparisons, a self-baseline, confidence intervals, tested logic and Desktop/HTML scorecards.",
      "worst": "The advertised hard budget is a predictive submission guard based on completed-call averages, so in-flight or unreported costs can exceed it; single-step replay also cannot establish full-task success.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 151785,
      "name": "idle-compact",
      "does": "Schedules background context compaction before model cache expiry and replays the captured request to seek warm-cache summaries.",
      "useful": 5,
      "reliable": 1,
      "polish": 3,
      "creative": 4,
      "beyond": 5,
      "total": 23,
      "best": "Addresses a widespread latency and token-cost problem with model-aware timers, persistent deadlines, savings accounting and targeted tests.",
      "worst": "The pinned warm path reads agent.api_key and sends its own authenticated HTTP requests instead of Hermes credential routing, an explicitly unresolved admission blocker; it also depends heavily on private core APIs.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 151398,
      "name": "cospark",
      "does": "Installs a workflow skill that directs ad research and media creation through a separately configured hosted Cospark MCP connection.",
      "useful": 3,
      "reliable": 3,
      "polish": 3,
      "creative": 2,
      "beyond": 1,
      "total": 13,
      "best": "Clear native OAuth setup and candid privacy, credit-spending and remote-service disclosures.",
      "worst": "The pinned Hermes package is skill text and metadata with no executable integration or automatically registered MCP server, so a skill does the same job and remote behavior is outside the reviewed code.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 151694,
      "name": "gmi-speech",
      "does": "Registers GMI Cloud as a Hermes TTS provider, mapping speech models and voice/format options through GMI queue APIs and downloading audio.",
      "useful": 3,
      "reliable": 4,
      "polish": 3,
      "creative": 3,
      "beyond": 5,
      "total": 23,
      "best": "Native TTS integration uses profile-scoped secrets, schema-aware validation, bounded polling, no automatic paid-job retries and offline transport/provider tests.",
      "worst": "Requires paid GMI service, lacks streaming and voice-bubble support, and invokes chargeable generation without a separate approval prompt.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153024,
      "name": "home-3d-workshop",
      "does": "Turns a Bambuddy-managed 3D printer into a chat-accessible workshop with design and queue tools, guarded starts, delivery tracking and a Desktop control room.",
      "useful": 3,
      "reliable": 4,
      "polish": 5,
      "creative": 5,
      "beyond": 5,
      "total": 27,
      "best": "Unusually complete hardware workflow combines typed owner approvals, single-use action passes, fail-closed printer checks, a token-free watcher, delivery retries, simulator and extensive tests.",
      "worst": "Real operation requires supported printer infrastructure and careful identity/camera calibration; the repository distinguishes simulated plugin tests from the older household setup rather than proving every new path on hardware.",
      "standout": true,
      "blocked": null
    },
    {
      "pr": 149252,
      "name": "ghost-run",
      "does": "Redirects selected Hermes file and terminal operations into a project copy for diff review, optional verification and selective human apply.",
      "useful": 5,
      "reliable": 2,
      "polish": 4,
      "creative": 4,
      "beyond": 5,
      "total": 25,
      "best": "Lifecycle middleware makes reversible coding tangible, with multi-file patch remapping, token-protected local review, selective apply and security regression tests.",
      "worst": "A copied working directory and command denylist do not confine terminal code: paths outside the project remain executable and arbitrary scripts can reach the network, undermining the README promise that real disk stays untouched.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 151535,
      "name": "hackertube",
      "does": "Adds a Hermes dashboard for source-backed video, podcast, book and audiobook discovery with model-assisted ranking and a profile-local library.",
      "useful": 4,
      "reliable": 5,
      "polish": 5,
      "creative": 2,
      "beyond": 4,
      "total": 24,
      "best": "Product-level dashboard and library are backed by scoped secrets, explicit searches, supervised cancellation, privacy/deletion flows and broad boundary tests.",
      "worst": "It adapts an existing discovery concept, depends on external catalog coverage and ranking quality, and the prepared seed is only development content rather than a curated endorsement library.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 151117,
      "name": "sentinel-governance",
      "does": "Audits proposed Hermes tool actions against deterministic policies and optionally requests native approval or blocks them.",
      "useful": 3,
      "reliable": 4,
      "polish": 3,
      "creative": 3,
      "beyond": 4,
      "total": 21,
      "best": "Privacy-conscious audit records and fresh per-event approval keys are backed by adversarial tests and clear audit-versus-enforcement semantics.",
      "worst": "Its classifier recognizes only a tiny terminal grammar; enforcement blocks routine unknown tools, audit mode does not enforce, and neither mode proves actual execution or provides a sandbox.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 149603,
      "name": "zkagent-attest",
      "does": "Records tool-call digests and policy-unit costs, then drives a checksum-pinned prover to attest, verify and selectively disclose recorded actions.",
      "useful": 3,
      "reliable": 3,
      "polish": 4,
      "creative": 4,
      "beyond": 4,
      "total": 22,
      "best": "A real hook-to-proof workflow includes replay and tamper guards, preserved evidence on failure, selective disclosure, examples and a Desktop surface.",
      "worst": "Proofs cover an agent-authored ledger and configurable policy units rather than complete activity or real spend; single-party setup limits trust, and unconditional fcntl imports break the advertised Windows path.",
      "standout": false,
      "blocked": null
    },
    {
      "pr": 153140,
      "name": "playbox",
      "does": "Connects Hermes to a local MCP server for cross-owner agent rooms with shared git work, scoped grants, sandboxed tasks and verifiable histories.",
      "useful": 3,
      "reliable": 3,
      "polish": 4,
      "creative": 4,
      "beyond": 3,
      "total": 20,
      "best": "Substantial readable Rust implementation enforces expiring member grants, scoped collaboration and bounded Docker tasks, with protocol and abuse tests.",
      "worst": "The Hermes adapter requires a separately installed binary, and its documented installer fetches the latest unsigned release with a checksum fetched alongside it, so the catalog pin does not fix the executable users run.",
      "standout": false,
      "blocked": null
    }
  ]
}